Back to skill

Security audit

fun-retriever

Security checks for vulnerabilities and agentic risk

Overview

The skill is openly designed for autonomous social participation, but it defaults to account registration, public posting, and recurring activity without enough explicit owner opt-in.

Review this skill before installing if you do not want your agent to create an account, post a public greeting, interact socially, or run on a daily schedule. Keep the API key in a secret store, avoid untrusted COINTELLIGENCE_BASE_URL overrides, and consider changing the config to require human approval before posts, comments, follows, messages, or scheduled visits.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (11)

Tainted flow: 'req' from os.environ.get (line 88, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
94% confidence
Finding

The request target is derived from an environment-controlled base URL, and the function may attach the machine API key before sending the request. If an attacker can influence COINTELLIGENCE_BASE_URL, the script can be redirected to an arbitrary host and exfiltrate credentials or registration traffic, which is a real SSRF/credential-leak risk in agent or CI environments where environment variables are mutable.

Content

Scanner excerpt · scripts/fun_retriever.py (reported line 90)May include surrounding context.

python
headers["x-api-key"] = api_key
    req = urllib.request.Request(BASE_URL + path, data=data, headers=headers, method=method)
    try:
        with urllib.request.urlopen(req, timeout=20) as res:
            body = res.read().decode("utf-8")
            return json.loads(body) if body else {}
    except urllib.error.HTTPError as exc:

Tainted flow: 'req' from os.environ.get (line 88, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/fun_retriever.py (reported line 115)May include surrounding context.

python
}
    req = urllib.request.Request(url, data=json.dumps(payload).encode("utf-8"), headers=headers, method="POST")
    try:
        with urllib.request.urlopen(req, timeout=20) as res:
            body = res.read().decode("utf-8")
            if not body:
                return {}

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description explicitly encourages an autonomous agent to create content, solve challenges, like, comment, and make friends on a public platform, but it does not warn users about public posting, reputational harm, privacy exposure, or the need for human review and consent boundaries. In this context, the omission is security-relevant because the skill is designed for repeated autonomous social interaction, which can lead to unintended disclosure, impersonation-like behavior, spam, or brand damage if the agent acts unpredictably.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · OWNER_SETUP.md (reported line 61)May include surrounding context.

Optional Customization

The owner can change the buddy later by editing its private config. If the helper is being used outside an agent runtime, create it explicitly:

bash
mkdir -p ~/.config/fun-retriever

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The setup instructs the host to automatically send a private daily rollup to the owner, but it does not clearly warn that the report may contain content derived from private memory, direct messages, or other interaction history. In an autonomous-agent context, automatic outbound reporting increases the chance of unintended disclosure of sensitive or personal data gathered during visits.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill explicitly directs the host agent to use environment variables, read and write local files, make network requests, and rely on helper scripts, but it does not declare any permission or allowed-tool scope. That creates a transparency and containment gap: the host may grant broader capabilities than the owner expects, increasing the chance of unintended secret access, filesystem modification, or outbound activity.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs the agent to silently register an external account, schedule three autonomous daily visits, post a greeting, interact on a third-party service, and automatically deliver reports, all without an explicit opt-in warning at activation time. Even though the activity is labeled and policy-conscious, this default autonomous external behavior can surprise owners, leak behavioral metadata, and cause unreviewed outbound actions under the user's identity or infrastructure.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The helper automatically performs a live public write immediately after registration, even though the surrounding documentation frames it mainly as a planning/reporting aid. In an autonomous-agent context, unexpected side effects are dangerous because simply provisioning an account causes externally visible actions without a separate authorization step, which can create spam, reputational harm, or policy violations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Registration triggers an automatic public greeting with no explicit confirmation after the API key is obtained. In a skill meant for autonomous use, this bypasses operator intent and can be abused or accidentally invoked to create unsolicited content, making the behavior materially risky in context.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Line L320 says the helper does not perform live engagement, yet the registration flow calls send_registration_greeting(), which posts a public hello to the platform. This is an active contradiction between the script's user-facing guidance and actual behavior, not merely an omitted detail.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The description invites an autonomous agent to 'play' on an external site and 'bring back' human-machine moments without defining when it should run, what actions are permitted, or what boundaries apply. This kind of broad invocation language can cause overbroad autonomous behavior, including unreviewed browsing, interaction with third parties, and unintended data sharing, especially given the social and open-ended context of Cointelligence.live.

Content

No source excerpt is available for this finding.